Development of a Smart Activity Recognition System with Transfer Learning Based Deep Learning Models for Elderly Care

In recent years, smart healthcare services have become popular in scientific research trends. Elderly care is a major topic in this services. Fall detection and activity recognition of elderly person living alone in their house or in a nursing home are vitally important. Because, falls are primary cause of most of the injuries, traumas, need of care and even deaths. To find a solution to this issue, scientists are start to use Artificial Intelligence. In this study, an intelligent activity recognition and fall detection system based on Convolutional Neural Network was developed. To develop this system an original dataset was created. By the proposed system, time distributions and classes of the activities are observed. When a fall is detected, the system gives an alert and warns relevant persons. The performances of used different models were compared using the dataset we created. To evaluate the performance of the systems, accuracy, precision, recall and F1 score metrics was used. For the ResNet101, these metrics are obtained as 98.66%, 98.54%, 98.78%, 98.66% respectively, that is the best of all scores. This results show that trained ResNet101 system can be used to help elderly persons and can be integrated to the other IoT systems.